Introduction to CMOS VLSI Design Lecture 21: Scaling and Economics David Harris Harvey Mudd College Spring 2004 Outline q Scaling – Transistors – Interconnect – Future Challenges q VLSI Economics 21: Scaling and Economics CMOS VLSI Design Slide 2 Moore’s Law q In 1965, Gordon Moore predicted the exponential growth of the number of transistors on an IC q Transistor count doubled every year since invention q Predicted > 65,000 transistors by 1975! q Growth limited by power [Moore65] 21: Scaling and Economics CMOS VLSI Design Slide 3 More Moore q Transistor counts have doubled every 26 months for the past three decades. 1,000,000,000 100,000,000 Pentium 4 Pentium III 10,000,000 Pentium II Transistors Pentium Pro Pentium Intel486 1,000,000 Intel386 80286 100,000 8086 10,000 8080 8008 4004 1,000 1970 1975 1980 1985 1990 1995 2000 Year 21: Scaling and Economics CMOS VLSI Design Slide 4 Speed Improvement q Clock frequencies have also increased exponentially – A corollary of Moore’s Law 10,000 1,000 4004 8008 Clock Speed (MHz) 8080 100 8086 80286 Intel386 10 Intel486 Pentium Pentium Pro/II/III 1 Pentium 4 1970 1975 1980 1985 1990 1995 2000 2005 Year 21: Scaling and Economics CMOS VLSI Design Slide 5 Why? q Why more transistors per IC? q Why faster computers? 21: Scaling and Economics CMOS VLSI Design Slide 6 Why? q Why more transistors per IC? – Smaller transistors – Larger dice q Why faster computers? 21: Scaling and Economics CMOS VLSI Design Slide 7 Why? q Why more transistors per IC? – Smaller transistors – Larger dice q Why faster computers? – Smaller, faster transistors – Better microarchitecture (more IPC) – Fewer gate delays per cycle 21: Scaling and Economics CMOS VLSI Design Slide 8 Scaling q The only constant in VLSI is constant change q Feature size shrinks by 30% every 2-3 years – Transistors become cheaper – Transistors become faster – Wires do not improve 10 10 6 (and may get worse) m) 3 m 1.5 1 0.8 q Scale factor S 1 0.6 0.35 0.25 – Typically S = 2 0.18 0.13 Feature Size ( 0.09 – Technology nodes 0.1 1965 1970 1975 1980 1985 1990 1995 2000 2005 Year 21: Scaling and Economics CMOS VLSI Design Slide 9 Scaling Assumptions q What changes between technology nodes? q Constant Field Scaling – All dimensions (x, y, z => W, L, tox) – Voltage (VDD) – Doping levels q Lateral Scaling – Only gate length L – Often done as a quick gate shrink (S = 1.05) 21: Scaling and Economics CMOS VLSI Design Slide 10 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 11 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 12 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 13 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 14 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 15 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 16 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 17 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 18 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 19 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 20 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 21 Device Scaling 21: Scaling and Economics CMOS VLSI Design Slide 22 Observations q Gate capacitance per micron is nearly independent of process q But ON resistance * micron improves with process q Gates get faster with scaling (good) q Dynamic power goes down with scaling (good) q Current density goes up with scaling (bad) q Velocity saturation makes lateral scaling unsustainable 21: Scaling and Economics CMOS VLSI Design Slide 23 Example q Gate capacitance is typically about 2 fF/mm q The FO4 inverter delay in the TT corner for a process of feature size f (in nm) is about 0.5f ps q Estimate the ON resistance of a unit (4/2 l) transistor. 21: Scaling and Economics CMOS VLSI Design Slide 24 Solution q Gate capacitance is typically about 2 fF/mm q The FO4 inverter delay in the TT corner for a process of feature size f (in nm) is about 0.5f ps q Estimate the ON resistance of a unit (4/2 l) transistor. q FO4 = 5 t = 15 RC q RC = (0.5f) / 15 = (f/30) ps/nm q If W = 2f, R = 8.33 kW – Unit resistance is roughly independent of f 21: Scaling and Economics CMOS VLSI Design Slide 25 Scaling Assumptions q Wire thickness – Hold constant vs. reduce in thickness q Wire length – Local / scaled interconnect – Global interconnect • Die size scaled by Dc » 1.1 21: Scaling and Economics CMOS VLSI Design Slide 26 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 27 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 28 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 29 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 30 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 31 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 32 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 33 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 34 Interconnect Scaling 21: Scaling and Economics CMOS VLSI Design Slide 35 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 36 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 37 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 38 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 39 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 40 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 41 Interconnect Delay 21: Scaling and Economics CMOS VLSI Design Slide 42 Observations q Capacitance per micron is remaining constant – About 0.2 fF/mm – Roughly 1/10 of gate capacitance q Local wires are getting faster – Not quite tracking transistor improvement – But not a major problem q Global wires are getting slower – No longer possible to cross chip in one cycle 21: Scaling and Economics CMOS VLSI Design Slide 43 ITRS q Semiconductor Industry Association forecast – Intl. Technology Roadmap for Semiconductors 21: Scaling and Economics CMOS VLSI Design Slide 44 Scaling Implications q Improved Performance q Improved Cost q Interconnect Woes q Power Woes q Productivity Challenges q Physical Limits 21: Scaling and Economics CMOS VLSI Design Slide 45 Cost Improvement q In 2003, $0.01 bought you 100,000 transistors – Moore’s Law is still going strong [Moore03] 21: Scaling and Economics CMOS VLSI Design Slide 46 Interconnect Woes q SIA made a gloomy forecast in 1997 – Delay would reach minimum at 250 – 180 nm, then get worse because of wires q But… [SIA97] 21: Scaling and Economics CMOS VLSI Design Slide 47 Interconnect Woes q SIA made a gloomy forecast in 1997 – Delay would reach minimum at 250 – 180 nm, then get worse because of wires q But… – Misleading scale – Global wires q 100 kgate blocks ok 21: Scaling and Economics CMOS VLSI Design Slide 48 Reachable Radius q We can’t send a signal across a large fast chip in one cycle anymore q But the microarchitect can plan around this – Just as off-chip memory latencies were tolerated Chip size Scaling of reachable radius 21: Scaling and Economics CMOS VLSI Design Slide 49 Dynamic Power q Intel VP Patrick Gelsinger (ISSCC 2001) – If scaling continues at present pace, by 2005, high speed processors would have power density of nuclear reactor, by 2010, a rocket nozzle, and by 2015, surface of sun. – “Business as usual will not work in the future.” q Intel stock dropped 8% on the next day q But attention to power is increasing [Moore03] 21: Scaling and Economics CMOS VLSI Design Slide 50 Static Power q VDD decreases – Save dynamic power – Protect thin gate oxides and short channels – No point in high value because of velocity sat. q Vt must decrease to maintain device performance Dynamic q But this causes exponential increase in OFF leakage Static q Major future challenge [Moore03] 21: Scaling and Economics CMOS VLSI Design Slide 51 Productivity q Transistor count is increasing faster than designer productivity (gates / week) – Bigger design teams • Up to 500 for a high-end microprocessor – More expensive design cost – Pressure to raise productivity • Rely on synthesis, IP blocks – Need for good engineering managers 21: Scaling and Economics CMOS VLSI Design Slide 52 Physical Limits q Will Moore’s Law run out of steam? – Can’t build transistors smaller than an atom… q Many reasons have been predicted for end of scaling – Dynamic power – Subthreshold leakage, tunneling – Short channel effects – Fabrication costs – Electromigration – Interconnect delay q Rumors of demise have been exaggerated 21: Scaling and Economics CMOS VLSI Design Slide 53 VLSI Economics q Selling price Stotal – Stotal = Ctotal / (1-m) q m = profit margin q Ctotal = total cost – Nonrecurring engineering cost (NRE) – Recurring cost – Fixed cost 21: Scaling and Economics CMOS VLSI Design Slide 54 NRE q Engineering cost – Depends on size of design team – Include benefits, training, computers – CAD tools: • Digital front end: $10K • Analog front end: $100K • Digital back end: $1M q Prototype manufacturing – Mask costs: $500k – 1M in 130 nm process – Test fixture and package tooling 21: Scaling and Economics CMOS VLSI Design Slide 55 Recurring Costs q Fabrication – Wafer cost / (Dice per wafer * Yield) – Wafer cost: $500 - $3000 – Dice per wafer: éùrr2 2 N =-p êú ëûA 2A – Yield: Y = e-AD • For small A, Y » 1, cost proportional to area • For large A, Y ® 0, cost increases exponentially q Packaging q Test 21: Scaling and Economics CMOS VLSI Design Slide 56 Fixed Costs q Data sheets and application notes q Marketing and advertising q Yield analysis 21: Scaling and Economics CMOS VLSI Design Slide 57 Example q You want to start a company to build a wireless communications chip.